Self-repairing hardware with astrocyte-neuron networks

摘要

A Self-rePAiring spiking Neural NEtwoRk (SPANNER) hardware architecture is presented in this paper. It is based on a software model of an astrocyte-neuron network which previously demonstrated the ability to self-detect faults and self-repair autonomously. Experimental results in this paper show that when faults occur at the synapse, remaining healthy synapses of the same neuron are enhanced by the feedback from the astrocyte, which enables system functionality to be maintained. This is the first time that astrocyte cells merged within spiking neurons demonstrate a self-repairing capability in hardware. This repair capability achieves a much more finegrained repair capability in hardware compared to the conventional fault tolerance techniques.